ABSTRACT Federated learning enables collaborative intrusion detection without centralizing sensitive network data, but existing approaches optimize solely for detection accuracy while neglecting critical performance and privacy constraints. This paper presents MOFL‐CyberNet, a novel multi‐objective federated learning framework that simultaneously optimizes three competing objectives: detection accuracy, network efficiency, and privacy preservation. The framework integrates three key innovations: (1) a lightweight Cross‐Attention Transformer architecture (5.0 M parameters) specifically designed for distributed intrusion detection, capturing complex attack patterns with minimal computational overhead; (2) an adaptive Pareto‐optimal aggregation mechanism using NSGA‐II that dynamically balances objectives based on real‐time network conditions; and (3) comprehensive privacy‐preserving mechanisms including differential privacy and secure aggregation. Through rigorous evaluation on four major data sets totaling over 27 million samples (NSL‐KDD, EDGE‐IIoTset, CICIDS‐2018, and CIC‐IoT‐2023), MOFL‐CyberNet demonstrates exceptional performance: 97.86% detection accuracy coupled with 47% latency reduction and 36% energy savings compared to state‐of‐the‐art baselines. These results demonstrate that multi‐objective optimization can achieve favorable performance across all evaluated metrics concurrently making federated intrusion detection practical for resource‐constrained IoT deployments.
Al-madani et al. (Wed,) studied this question.